Automatic vs Manual Provenance Abstractions: Mind the Gap

TL;DR

This paper compares manual and semi-automatic provenance abstraction techniques in scientific workflows, revealing significant differences in data artefacts retained and discussing implications for future research.

Contribution

It provides an empirical comparison between manual and semi-automatic provenance abstractions, highlighting their overlaps and differences in data retention.

Findings

Semi-automatic and manual abstractions largely overlap process-wise.

Significant mismatch in data artefacts retained between approaches.

Discussion on reasons and future research directions.

Abstract

In recent years the need to simplify or to hide sensitive information in provenance has given way to research on provenance abstraction. In the context of scientific workflows, existing research provides techniques to semi automatically create abstractions of a given workflow description, which is in turn used as filters over the workflow's provenance traces. An alternative approach that is commonly adopted by scientists is to build workflows with abstractions embedded into the workflow's design, such as using sub-workflows. This paper reports on the comparison of manual versus semi-automated approaches in a context where result abstractions are used to filter report-worthy results of computational scientific analyses. Specifically; we take a real-world workflow containing user-created design abstractions and compare these with abstractions created by ZOOM UserViews and Workflow Summaries systems. Our comparison shows that semi-automatic and manual approaches largely overlap from a process perspective, meanwhile, there is a dramatic mismatch in terms of data artefacts retained in an abstracted account of derivation. We discuss reasons and suggest future research directions.